This project applies advanced econometric methods to explore the intersection of energy consumption and vehicle electrification in the United States through time series forecasting. As the transportation sector undergoes a rapid shift toward electric vehicles (EVs), understanding the broader energy implications of this transition is critical for infrastructure planning, environmental policy, and market strategy.
To examine these trends, I analyzed three key datasets:
Monthly natural gas consumption data — specifically deliveries to the electric power sector — sourced from the Federal Reserve Economic Data (FRED)
Monthly Plug-in Electric Vehicle (PEV) sales from Argonne National Laboratory
Monthly Hybrid Electric Vehicle (HEV) sales, also from Argonne National Laboratory
These datasets span multiple years and extend through early 2025, offering recent and high-resolution insights into energy and vehicle market behavior. Natural gas was chosen as a proxy for electricity generation, given its dominant role in powering the U.S. grid, while HEVs and PEVs were selected as representative alternatives in the shift away from internal combustion engines.
The economic implications of this research are significant, potentially informing policy decisions related to energy infrastructure investment,and incentive structures for EV adoption. By forecasting trends in these three time series, the goal is to better understand how EV adoption may reshape both transportation and energy landscapes in the near future, with particular attention to price elasticities, market equilibrium dynamics, and welfare effects.
## HEV Training set size: 137 observations
## HEV Test set size: 35 observations
## PEV Training set size: 137 observations
## PEV Test set size: 35 observations
## Natural Gas Training set size: 241 observations
## Natural Gas Test set size: 61 observations
Here, I present the STL decomposition along with monthly sales and consumption trends, accompanied by a brief analysis of the underlying patterns
Natural Gas Consumption (BCF)
Long-term Trends: There is a strong upward trend in natural gas consumption from 2000 to 2025, especially pronounced after 2010. This suggests growing dependence on natural gas, likely due to the transition away from coal in the power sector.
Seasonality: Clear, sharp annual seasonal cycles are present, with peaks occurring consistently in the winter months. This aligns with seasonal heating demand.
STL Decomposition Insight: The trend component confirms a steady rise over time. The seasonal component is highly regular and strong, while the remainder component is relatively stable, indicating a well-behaved and predictable series.
Hybrid Electric Vehicle (HEV) Sales
Long-term Trends: HEV sales exhibit moderate growth between 2011 and 2016, followed by a plateau and some fluctuation from 2017–2020. A sharp increase occurs post-2021, suggesting renewed consumer interest—possibly due to higher gas prices or policy incentives.
Seasonality: Seasonality is present but less pronounced compared to natural gas. Fluctuations are more irregular, with notable dips likely tied to economic or supply-chain shocks (e.g., COVID in 2020).
STL Decomposition Insight: The trend shows relatively flat sales through the mid-2010s, with a steep rise beginning around 2021. The seasonal component is present but modest, while the remainder shows noise and some large outliers, indicating irregular influences.
Plug-in Electric Vehicle (PEV) Sales
Long-term Trends: EV sales have grown steadily since 2011, with a major surge beginning in 2021 and continuing through 2025. This sharp growth reflects increased EV adoption, federal/state incentives, and a broader market shift toward electrification.
Seasonality: Compared to HEVs, PEVs show a stronger seasonal pattern with more regular intra-year fluctuations—likely related to production cycles and end-of-year consumer purchases.
STL Decomposition Insight: The trend is clearly upward, accelerating post-2021. Seasonality is evident and increasing in amplitude as total sales grow. The remainder component shows more pronounced deviations in recent years, possibly reflecting supply chain volatility or changing consumer demand.
Key Takeaways
All three time series show clear upward trends, particularly from 2020 onward, aligning with broader energy and transportation transitions.
Natural gas consumption displays the strongest and most consistent seasonal pattern, indicative of weather-driven demand.
HEV sales show a delayed and flatter trend, but post-pandemic dynamics have driven a sharp resurgence.
PEV sales are accelerating most rapidly, with growing seasonal variation and an upward trend that suggests strong adoption momentum.
STL decompositions confirm that while all three series contain seasonal and trend components, natural gas is the most predictable, whereas PEV and HEV sales are more susceptible to structural shifts and external shocks.
A log transformation is applied to the three time series to stabilize variance and improve interpretability.
As shown, the log transformation effectively stabilized the variance in the data. Therefore, we will apply this transformation in our modeling process and back-transform the forecasts to the original scale to ensure interpretability.
## [1] "Seasonal differences needed for natural gas: 1"
## [1] "Regular differences needed after seasonal differencing for natural gas: 0"
## [1] "Seasonal differences needed for PEV sales: 0"
## [1] "Regular differences needed for PEV sales: 1"
## [1] "Seasonal differences needed for HEV sales: 0"
## [1] "Regular differences needed for HEV sales: 0"
Based on unit root and KPSS tests, the log-transformed data for all series was assessed for stationarity. The following differencing was applied to achieve stationarity:
Natural Gas: 1 seasonal difference (lag = 12), no regular differencing required.
PEV Sales: No seasonal differencing, 1 regular difference (lag = 1).
HEV Sales: No seasonal or regular differencing required.
Based on our model evaluation and ACF/PACF diagnostics, we initially estimated models guided by the ARIMA structure. However, the ARIMAX models for PEV and HEV sales, and the automated ARIMA model for natural gas consumption, demonstrated superior performance. Therefore, we proceed by presenting and using only these models in our analysis.
## [1] "Accuracy for Natural Gas (AUTO ARIMA):"
## # A tibble: 1 × 10
## .model .type ME RMSE MAE MPE MAPE MASE RMSSE ACF1
## <chr> <chr> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
## 1 AUTO_NG Test 0.0145 0.0474 0.0394 0.186 0.500 NaN NaN 0.428
## [1] "Accuracy for PEV Sales (ARIMAX):"
## # A tibble: 1 × 10
## .model .type ME RMSE MAE MPE MAPE MASE RMSSE ACF1
## <chr> <chr> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
## 1 arimax_PEV Test -0.0978 0.261 0.179 -1.03 1.78 NaN NaN 0.664
## [1] "Accuracy for HEV Sales (ARIMAX):"
## # A tibble: 1 × 10
## .model .type ME RMSE MAE MPE MAPE MASE RMSSE ACF1
## <chr> <chr> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
## 1 arimax_HEV Test 0.263 0.403 0.336 2.35 3.09 NaN NaN 0.775
While the models demonstrate solid performance based on key statistical metrics, we will later visualize their forecasting accuracy on the test data and report the percentage accuracy of forecasted versus actual values as part of our outlook report. During model development, we tested multiple iterations of the ARIMAX models using various combinations of exogenous variables. Ultimately, we found that including HEV and PEV sales as predictors of each other produced the best-performing models. Initially, we experimented with incorporating all available variables from our dataset as exogenous inputs, but the core accuracy metrics remained unchanged. As a result, we excluded variables that did not contribute meaningfully to the model’s predictive power.
We now examine the residuals of each model and perform the Ljung-Box test to assess whether the residuals resemble white noise, thereby confirming the adequacy of the model fit.
## Ljung-Box Test (lag = 12):
## Natural Gas Residuals:
## p-value = 0.243 - Residuals resemble white noise.
## PEV Residuals:
## p-value = 0.1324 - Residuals resemble white noise.
## HEV Residuals:
## p-value = 0.1393 - Residuals resemble white noise.
All p-values are above the 0.05 threshold!
## 12-Month Forecast Accuracy:
## Natural Gas:
## MAPE = 0.5 %
## Approximate Forecast Accuracy = 99.5 %
## PEV Sales:
## MAPE = 1.78 %
## Approximate Forecast Accuracy = 98.22 %
## HEV Sales:
## MAPE = 3.09 %
## Approximate Forecast Accuracy = 96.91 %
Based on our testing data validation using an 80/20 training-test split, the models demonstrate strong forecast performance over a 12-month horizon:
Natural Gas: MAPE = 0.5%, Approx. Forecast Accuracy = 99.5%
PEV Sales: MAPE = 1.78%, Approx. Forecast Accuracy = 98.22%
HEV Sales: MAPE = 3.09%, Approx. Forecast Accuracy = 96.91%
These results indicate that our models are performing well beyond acceptable thresholds, with consistently high accuracy across all series. We now proceed to generate out-of-sample forecasts for the next 12 months using the full dataset.
These forecasts provide useful insights for energy planners, automotive strategists, and policy agencies aiming to guide the transition toward cleaner transportation and better anticipate fuel demand cycles.
Our forecasts of natural gas consumption and electric vehicle adoption trends reveal significant economic implications for the next 12 months:
The forecast data suggests an ongoing structural shift in the U.S. energy and transportation sectors:
Natural Gas Consumption: Our model projects continued seasonal patterns with moderate year-over-year growth, reflecting natural gas’s sustained role in electricity generation despite renewable energy expansion. The stability in this forecast suggests that infrastructure planning for natural gas delivery can proceed with reasonable certainty.
Plug-in Electric Vehicle (PEV) Sales: The strong growth trajectory indicates accelerating market penetration, reflecting both consumer preferences and policy effectiveness. This trend appears resilient and suggests the EV market may be approaching a tipping point where network effects (charging infrastructure, model availability) begin to self-reinforce.
Hybrid Electric Vehicle (HEV) Sales: The projected decline suggests a potential substitution effect, where consumers are increasingly bypassing hybrid technology in favor of fully electric options. This pattern reflects a market maturation where transitional technologies face pressure as the end-state technology becomes more accessible.
Grid Planning: The consistent natural gas consumption forecast indicates utilities should maintain current capacity planning while preparing for seasonal demand fluctuations. However, the accelerating growth in PEV adoption signals a need to prepare for increased electricity demand, particularly in residential areas during evening charging periods.
Infrastructure Investment: Our forecasts support continued investment in electricity distribution infrastructure, particularly in high-EV-adoption regions, to prevent localized grid constraints as PEV market share grows.
EV Incentive Design: The diverging trajectories between PEVs (growing) and HEVs (declining) suggest that policy incentives could be recalibrated to focus more on fully electric vehicles and less on hybrid technology, potentially accelerating the transition to zero-emission transportation.
Charging Infrastructure: The robust PEV growth forecast justifies accelerated public investment in charging stations, particularly along interstate corridors and in multi-unit dwelling areas where private investment may lag.
Emissions Planning: The forecasted growth in PEV adoption suggests potential acceleration in transportation sector decarbonization, which could inform updated emissions reduction timelines and carbon budget allocations.
Air Quality Modeling: Local air quality agencies can use these forecasts to anticipate pollution reduction benefits in urban areas with high projected EV adoption rates.
Automotive Manufacturers: The projected decline in HEV sales coupled with PEV growth suggests manufacturers should accelerate the transition of production capacity from hybrid to fully electric platforms.
Energy Companies: Natural gas providers can expect consistent demand from the electricity sector, while electricity providers should prepare for increased load from vehicle charging, particularly during evening hours.
Investors: These forecasts indicate continued growth opportunities in EV manufacturing, battery technology, and charging infrastructure, while suggesting caution regarding long-term investments in hybrid-specific technologies.
This economic outlook analysis would be particularly valuable to:
By providing quantitative forecasts with clear confidence intervals, this analysis helps these stakeholders develop data-driven strategies and policies to manage the ongoing energy transition, balance infrastructure needs, and optimize incentive structures for maximum environmental and economic benefit.
## NATURAL GAS FORECAST
## 3 months ahead (May 2025): 7.77 → 2,368 BCF
## 6 months ahead (August 2025): 7.902 → 2,702 BCF
## 9 months ahead (November 2025): 7.958 → 2,858 BCF
##
## PEV SALES FORECAST
## 3 months ahead (June 2025): 10.753 → 46,779 vehicles
## 6 months ahead (September 2025): 10.903 → 54,338 vehicles
## 9 months ahead (December 2025): 11.048 → 62,812 vehicles
##
## HEV SALES FORECAST
## 3 months ahead (June 2025): 11.255 → 77,229 vehicles
## 6 months ahead (September 2025): 11.131 → 68,255 vehicles
## 9 months ahead (December 2025): 11.088 → 65,400 vehicles